Research map / Video / Video segmentation and tracking
MixForensics: Blend Before You Encode for Generalizable AI-Generated Video Detection
Ruiqi Liu, ZiAn Wang, Ruoxin Chen, Xuehai Bai, Manni Cui, Ziheng Qin, Zhiyuan Yan, Shu Wu, Wenhao Wang
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes MixForensics, a generator-agnostic AI-generated video detector that blends frames before encoding rather than encoding then fusing, to better exploit multi-frame forensic evidence.
Problem
Detecting AI-generated videos in a generator-agnostic way; existing encode-then-fuse detectors fail to gain much from additional frames, so multi-frame evidence is underused.
Approach
Blend frames before encoding (blend-then-encode) instead of encoding each frame and fusing afterward, so the encoder can capture cross-frame forensic evidence that encode-then-fuse discards.
Key results
- Increasing frame budget from 1 to 8 under encode-then-fuse yields only a marginal gain for a voting-based image detector (abstract truncated before exact numbers).
Surprising findings
- Under the encode-then-fuse paradigm, increasing frames from 1 to 8 improves a voting-based image detector only slightly, despite the intuition that more frames offer richer evidence.
Limitations
- Full text unavailable; experimental details unknown from abstract alone.
Closest papers by meaning
- Generalizing Deepfake Video Detection with Plug-and-Play: Video-Level Blending and Spatiotemporal Adapter Tuning (CVPR 2025 · similarity 0.88)
- Seeing What Matters: Generalizable AI-generated Video Detection with Forensic-Oriented Augmentation (NeurIPS 2025 · similarity 0.87)
- BusterX: MLLM-Powered AI-Generated Video Forgery Detection and Explanation (NeurIPS 2026 · similarity 0.86)
- Community Forensics: Using Thousands of Generators to Train Fake Image Detectors (CVPR 2025 · similarity 0.86)
- IV-mixed Sampler: Leveraging Image Diffusion Models for Enhanced Video Synthesis (ICLR 2025 · similarity 0.86)
- Rethinking the Readout: Unlocking Video Backbones for AI-Generated Video Detection (NeurIPS 2026 · similarity 0.86)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
